1M-P-Bit Probabilistic Computer: What Founders Should Know
20 Jul 2026
The scale-up, by the numbers
A team of researchers has built what they describe as the largest probabilistic computer to date — a machine with 1,000,000 p-bits spread across 18 field-programmable gate arrays (FPGAs). According to the report, the system is capable of more than a trillion flips per second. The findings were detailed in a preprint posted to ArXiv on June 24, 2024.
Probabilistic computers, or p-computers, use "p-bits" — probabilistic bits that fluctuate between states — rather than the deterministic bits of classical computing or the qubits of quantum computing. This latest machine represents a dramatic jump in scale for the category.
From 8 to 1 million in five years
The timeline of progress is striking:
- 2019 — Scientists developed a probabilistic computer with just 8 p-bits.
- 2023 — Researchers scaled that up to a machine with 7,200 p-bits.
- 2024 — The newly detailed system jumps to 1,000,000 p-bits, a roughly 139x increase in a single year and a five-year progression from single digits to seven figures.
According to the report, Navid Anjum Aadit, a postdoctoral scholar in electrical engineering at Stanford University, said the machine communicates without global lockstep synchronization — a design detail tied to how the system scales across its 18 FPGAs.
Kerem Çamsarı, an associate professor of electrical and computer engineering at UC Santa Barbara, stated that probabilistic computers are not hardwired for a single problem but are programmable, general-purpose machines — a distinction from narrower, single-task accelerators.
What's still unknown
The report flags several open questions. The specific problems or applications the 1-million-p-bit machine has been tested on are not described. There's no information on the system's cost, power consumption, or physical size. No comparison is given against classical or quantum computing performance on equivalent tasks, and the affiliations or funding behind the project aren't detailed beyond the researchers' academic titles.
Perhaps most importantly for anyone evaluating the claims: it remains unclear whether the ArXiv preprint has undergone peer review or been published in a journal. As the report notes, findings based on a preprint may not yet be peer-reviewed and could be revised. Scaling from thousands to a million p-bits may also introduce engineering challenges that haven't been publicly addressed yet.
Why founders should care
For early-stage founders tracking novel compute architectures, this development is worth watching rather than acting on immediately:
- The rapid scale-up — from 8 p-bits in 2019 to 1 million in 2024 — may suggest that probabilistic computing hardware is maturing faster than expected, which could plausibly open new avenues in optimization and probabilistic computing applications over time.
- The described general-purpose programmability, if it holds up, could signal potential future use cases beyond narrow, single-problem accelerators — though this remains unconfirmed pending more detail on real-world testing.
- Because the underlying work is currently only a preprint, founders building in adjacent hardware or optimization spaces may want to hold off on strategic bets until peer-reviewed validation emerges, given the missing context around cost, power consumption, and comparative performance.
In short: an interesting scaling signal, but one still short on the commercial specifics — testing results, cost, and independent validation — that would let founders assess near-term relevance with confidence.